1 paper · 1 filter
Yiyuan Luo, Vaggos Chatziafratis
Text embedding models trained with contrastive learning learn to follow task instructions from instruction-paired data, while instruction-tuned LLMs already know how to follow them…